Difference between revisions of "Git"

From ArchWiki
Jump to navigation Jump to search
(Sorry, I couldn't wait for the big day :))
Line 9: Line 9:
{{Related|Concurrent Versions System}}
{{Related|Concurrent Versions System}}
{{Related articles end}}
{{Related articles end}}
{{Quote|I've met people who thought that git is a front-end for Github. They were wrong, git is a front-end for AUR.|Linus Torvalds}}
[[wikipedia:Git (software)|Git]] is the version control system (VCS) designed and developed by Linus Torvalds, the creator of the Linux kernel. Git is now used to maintain sources for the Linux kernel, as well as many other projects, including a number of Arch Linux projects.  
[[wikipedia:Git (software)|Git]] is the version control system (VCS) designed and developed by Linus Torvalds, the creator of the Linux kernel. Git is now used to maintain sources for the Linux kernel, as well as many other projects, including a number of Arch Linux projects.  

Revision as of 01:37, 31 March 2016


I've met people who thought that git is a front-end for Github. They were wrong, git is a front-end for AUR.

— Linus Torvalds

Git is the version control system (VCS) designed and developed by Linus Torvalds, the creator of the Linux kernel. Git is now used to maintain sources for the Linux kernel, as well as many other projects, including a number of Arch Linux projects.


Install the git package. For the development version, install git-gitAUR from the AUR. Check the optional dependencies when using tools such as git svn, git gui and gitk.

Note: To enable spell checking in git-gui, aspell is required, along with the dictionary corresponding to the LC_MESSAGES environment variable. See FS#28181.


In order to use Git you need to set at least a name and email:

$ git config --global user.name  "John Doe"
$ git config --global user.email "johndoe@foobar.com"

See #Tips and tricks for more settings.


This tutorial teaches how to use Git for basic distributed revision control of a project.

Git is a distributed version control system, which means that the entire history of changes to a repository is stored locally, in a directory called ./.git in the project directory. The project files which are visible to the user constitute the working tree. These files can be updated to match revisions stored in ./.git using git commands (e.g. git checkout), and new revisions can be created in turn by editing these files and running the appropriate git commands (e.g. git commit).

See man gitglossary for more complete definitions of the terms used in this tutorial.

A typical Git workflow is generally:

  1. Create a new project or clone a remote one.
  2. Create a branch to make changes; then commit those changes.
  3. Consolidate commits for better organization/understanding.
  4. Merge commits back into the main branch.
  5. (Optional) Push changes to a remote server.

Local repository

Staging operations

Initialize a new repository. This creates and initializes ./.git:

$ cd (your project)/
$ git init

To record the changes to the repository, they must first be added to the index, or staging area, with an operation often also referred to as staging. When you commit changes using git commit, it is the contents of the index which are committed to the current branch, not the contents of the working tree.

Note: The index is actually a binary file .git/index, which can be queried with git ls-files --stage.

To add files to the index from the working tree:

$ git add file1 file2

This records the current state of the files. If the files are subsequently modified, you can run git add again to "add" the new versions to the index.

Add all files:

$ git add .
Tip: To ignore some files from, e.g. git add ., create a .gitignore file (or files):
# File I'll likely delete

# Ignore all .html files, except 'important.html'

# Ignore all files recursively in 'DoNotInclude'

See gitignore(5) for details.

Remove a file from staging (--cached preserves the actual file(s)):

$ git rm (--cached) file

Remove all files:

$ git rm --cached -r .


$ git reset HEAD -- .

Here, HEAD is a "symbolic reference" to the current revision.

Rename a file:

$ git mv file1 file2
Note: git mv is a convenience command, roughly equivalent to mv file1 file2 followed by git rm file1 and git add file2. Rename detection during merges is based only on content similarity, and ignores the actual commands used to rename files. The broader rationale for this approach, which distinguishes Git from patch-based systems like Darcs, can be found here.

List files:

$ git ls-files

By default this shows files in the staging area (--cached files).

Commit changes

Once the content to be recorded is staged, commit them with:

$ git commit -m "First commit."

The -m (--message) option is for a short message: if omitted, the preset editor will be spawned to allow entering a longer message.

  • Always commit small changes frequently and with meaningful messages.
  • To add all the modified files to the index, and commit them in a single command (-a stands for --all):
$ git commit -am "First commit."

Edit the commit message for the last commit. This also amends the commit with any files which have been newly staged:

$ git commit --amend -m "Message."

Many of the commands in this article take commits as arguments. A commit can be identified by any of the following:

  • Its 40-digit SHA-1 hash (the first 7 digits are usually sufficient to identify it uniquely)
  • Any commit label such as a branch or tag name
  • The label HEAD always refers to the currently checked-out commit (usually the head of the branch, unless you used git checkout to jump back in history to an old commit)
  • Any of the above plus ~ to refer to previous commits. For example, HEAD~ refers to one commit before HEAD and HEAD~5 refers to five commits before HEAD.

View changes

Show differences between commits:

$ git diff HEAD HEAD~3

or between staging area and working tree:

$ git diff

Get a general overview of the changes:

$ git status

View history of changes (where "-N" is the number of latest commits):

$ git log -p (-N)

Branch a repository

Fixes and new features are usually tested in branches. When changes are satisfactory they can merged back into the default (master) branch.

Create a branch, whose name accurately reflects its purpose:

$ git branch help-section-addition

List branches:

$ git branch

Switch branches:

$ git checkout branch

Create and switch:

$ git checkout -b branch

Merge a branch back to the master branch:

$ git checkout master
$ git merge branch

The changes will be merged if they do not conflict. Otherwise, Git will print an error message, and annotate files in the working tree to record the conflicts. The annotations can be displayed with git diff. Conflicts are resolved by editing the files to remove the annotations, and committing the final version. See #Dealing with merges below.

When done with a branch, delete it with:

$ git branch -d branch


Adopting a good etiquette

  • When considering contributing to an existing project, read and understand its license, as it may excessively limit your ability to change the code. Some licenses can generate disputes over the ownership of the code.
  • Think about the project's community and how well you can fit into it. To get a feeling of the direction of the project, read any documentation and even the log of the repository.
  • When requesting to pull a commit, or submit a patch, keep it small and well documented; this will help the maintainers understand your changes and decide whether to merge them or ask you to make some amendments.
  • If a contribution is rejected, do not get discouraged, it is their project after all. If it is important, discuss the reasoning for the contribution as clearly and as patiently as possible: with such an approach a resolution may eventually be possible.

Clone a repository

To begin contributing to a project, clone its repository:

$ git clone location folder

location can be either a path or network address. Also, when cloning is done, the location is recorded so just a git pull will be needed later.

Pull requests

After making and committing some changes, the contributor can ask the original author to merge them. This is called a pull request.

To pull:

$ git pull location master

The pull command combines both fetching and merging. If there are conflicts (e.g. the original author made changes in the same time span), then it will be necessary to manually fix them.

Alternatively, the original author can pick the changes wanting to be incorporated. Using the fetch option (and log option with a special FETCH_HEAD symbol), the contents of the pull request can be viewed before deciding what to do:

$ git fetch location master
$ git log -p HEAD..FETCH_HEAD
$ git merge location master

Using remotes

Remotes are aliases for tracked remote repositories. A label is created defining a location. These labels are used to identify frequently accessed repositories.

Create a remote:

$ git remote add label location

Fetch a remote:

$ git fetch label

Show differences between master and a remote master:

$ git log -p master..label/master

View remotes for the current repository:

$ git remote -v

When defining a remote that is a parent of the fork (the project lead), it is defined as upstream.

Push to a repository

After being given rights from the original authors, push changes with:

$ git push location branch

When git clone is performed, it records the original location and gives it a remote name of origin.

So what typically is done is this:

$ git push origin master

If the -u (--set-upstream-to) option is used, the location is recorded so the next time just a git push is necessary.

Dealing with merges

See Basic Merge Conflicts in the Git Book for a detailed explanation on how to resolve merge conflicts. Merges are generally reversible. If wanting to back out of a merge one can usually use the --abort command (e.g. git merge --abort or git pull --abort).

Send patch to mailing list

If you want to send patches directly to a mailing list, you have to install the following packages: perl-authen-sasl, perl-net-smtp-ssl and perl-mime-tools.

Make sure you have configured you username and e-mail address, see #Configuration.

Configure your e-mail settings:

$ git config --global sendemail.smtpserver smtp.gmail.com
$ git config --global sendemail.smtpserverport 587
$ git config --global sendemail.smtpencryption tls
$ git config --global sendemail.smtpuser foobar@gmail.com

Now you should be able to send the patch to the mailing list (see also OpenEmbedded:How to submit a patch to OpenEmbedded#Sending patches):

$ git add filename
$ git commit -s
$ git send-email --to=openembedded-core@lists.openembedded.org --confirm=always -M -1

History and versioning

Searching the history

git log will give the history with a commit checksum, author, date, and the short message. The checksum is the "object name" of a commit object, typically a 40-character SHA-1 hash.

For history with a long message (where the "checksum" can be truncated, as long as it is unique):

$ git show (checksum)

Search for pattern in tracked files:

$ git grep pattern

Search in .c and .h files:

$ git grep pattern -- '*.[ch]'

Versioning for release

Tag commits for versioning:

$ git tag 2.14 checksum

Tagging is generally done for releasing/versioning but it can be any string. Generally annotated tags are used, because they get added to the Git database.

Tag the current commit with:

$ git tag -a 2.14 -m "Version 2.14"

List tags:

$ git tag -l

Delete a tag:

$ git tag -d 2.08

Update remote tags:

$ git push --tags

Organizing commits

Before submitting a pull request it may be desirable to consolidate/organize the commits. This is done with the git rebase --interactive option:

$ git rebase -i checksum

This will open the editor with a summary of all the commits in the range specified; in this case including the newest (HEAD) back to, but excluding, commit checksum. Or to use a number notation, use for example HEAD~3, which will rebase the last three commits:

pick d146cc7 Mountpoint test.
pick 4f47712 Explain -o option in readme.
pick 8a4d479 Rename documentation.

Editing the action in the first column will dictate how the rebase will be done. The options are:

  • pick — Apply this commit as is (the default).
  • edit — Edit files and/or commit message.
  • reword — Edit commit message.
  • squash — Merge/fold into previous commit.
  • fixup — Merge/fold into previous commit discarding its message.

The commits can be re-ordered or erased from the history (but be very careful with these). After editing the file, Git will perform the specified actions; if prompted to resolve merge problems, fix them and continue with git rebase --continue or back out with the git rebase --abort command.

Note: Squashing commits is only used for local commits, it will cause troubles on a repository that is shared by other people.

Tips and tricks

Using git-config

Git reads its configuration from a few INI-type configuration files:

  • Each repository contains a .git/config file for specific configuration.
  • Each user has a $HOME/.gitconfig file for fallback values.
  • /etc/gitconfig is used for system-wide defaults.

These files can be edited directly, but the usual method is to use git config, as shown in the examples below.

List the currently set variables:

$ git config {--local,--global,--system} --list

Set the default editor from vim to nano:

$ git config --global core.editor "nano -w"

Set the default push action:

$ git config --global push.default simple

Set a different tool for git difftool (meld by default):

$ git config --global diff.tool vimdiff

See git-config(1) and Git Configuration for more information.

Speeding up authentication

You may wish to avoid the hassle of authenticating interactively at every push to the Git server.

Protocol Defaults

If you are running a multiplexed SSH connection as shown above, Git over SSH might be faster than HTTPS. Also, some servers (like the AUR) only allow pushing via SSH. For example, the following config will set Git over SSH for any repository hosted on the AUR.

[url "ssh://aur@aur.archlinux.org/"]
	insteadOf = https://aur.archlinux.org/
	insteadOf = http://aur.archlinux.org/
	insteadOf = git://aur.archlinux.org/

Bash completion

In order to enable Bash completion, source /usr/share/git/completion/git-completion.bash in a Bash startup file. Alternatively, install bash-completion.

Git prompt

The Git package comes with a prompt script. To enable it, source the /usr/share/git/completion/git-prompt.sh script in a shell startup file, then set a custom prompt with the %s parameter:

  • For Bash: PS1='[\u@\h \W$(__git_ps1 " (%s)")]\$ '
  • For zsh: PS1='[%n@%m %c$(__git_ps1 " (%s)")]\$ '

When changing to a directory of a Git repository, the prompt will change to show the branch name. Extra details can be set to be shown by the prompt:

Shell variable Information
GIT_PS1_SHOWDIRTYSTATE + for staged, * if unstaged.
GIT_PS1_SHOWSTASHSTATE $ if something is stashed.
GIT_PS1_SHOWUNTRACKEDFILES % if there are untracked files.
GIT_PS1_SHOWUPSTREAM <,>,<> behind, ahead, or diverged from upstream.

GIT_PS1_SHOWUPSTREAM will need to be set to auto for changes to take effect.

Note: If you experience that $(__git_ps1) returns ((unknown)), then there is a .git folder in your current directory which does not contain any repository, and therefore Git does not recognize it. This can, for example, happen if you mistake Git's configuration file to be ~/.git/config instead of ~/.gitconfig.

Visual representation

To get an idea of the amount of work done:

$ git diff --stat

git log with forking representation:

$ git log --graph --oneline --decorate

git log graph alias (i.e. git graph will show a decorated version):

$ git config --global alias.graph 'log --graph --oneline --decorate'

Commit tips

Reset to previous commit (very dangerous, erases everything to specified commit):

$ git reset --hard HEAD^

If a repository address gets changed, its remote location will need to be updated:

$ git remote set-url origin git@address:user/repo.git

Signed-off-by line append (a name-email signature is added to the commit which is required by some projects):

 $ git commit -s

Signed-off-by automatically append to patches (when using git format-patch commit):

$ git config --local format.signoff true

Commit specific parts of files that have changed. This is useful if there are a large number of changes made that would be best split into several commits:

$ git add -p

Working with a non-master branch

Occasionally a maintainer will ask that work be done on a branch. These branches are often called devel or testing. Begin by cloning the repository.

To enter another branch beside master (git clone only shows master branch but others still exist, git branch -a to show):

$ git checkout -b branch origin/branch

Now edit normally; however to keep the repository tree in sync be sure to use both:

$ git pull --all
$ git push --all

Git server

How to set up connecting to repositories using varying protocols.


To use the SSH protocol, first set up a public SSH key; for that follow the guide at SSH keys. To set up a SSH server, follow the SSH guide.

With SSH working and a key generated, paste the contents of ~/.ssh/id_rsa.pub to ~/.ssh/authorized_keys (be sure it is all on one line). Now the Git repository can be accessed with SSH by doing:

$ git clone user@foobar.com:my_repository.git

You should now get an SSH yes/no question, if you have the SSH client setting StrictHostKeyChecking set to ask (the default). Type yes followed by Enter. Then you should have your repository checked out. Because this is with SSH, you also have commit rights now.

To modify an existing repository to use SSH, the remote location will need to be redefined:

$ git remote set-url origin git@localhost:my_repository.git

Connecting on a port other than 22 can be configured on a per-host basis in /etc/ssh/ssh_config or ~/.ssh/config. To set up ports for a repository, if the repository is in ~/ and using 443 for the port:

[remote "origin"]
    url = ssh://user@foobar.com:443/~my_repository/repo.git

Smart HTTP

Git is able to use the HTTP(S) protocol as efficiently as the SSH or Git protocols, by utilizing the git-http-backend. Furthermore it is not only possible to clone or pull from repositories, but also to push into repositories over HTTP(S).

The setup for this is rather simple as all you need to have installed is the Apache web server (apache, with mod_cgi, mod_alias, and mod_env enabled) and of course, git.

Once you have your basic setup running, add the following to your Apache configuration file, which is usually located at:

<Directory "/usr/lib/git-core*">
    Require all granted
SetEnv GIT_PROJECT_ROOT /srv/git
ScriptAlias /git/ /usr/lib/git-core/git-http-backend/

This assumes your Git repositories are located at /srv/git and that you want to access them via something like: http(s)://your_address.tld/git/your_repo.git.

Note: Make sure that Apache can read and write to your repositories.

For more detailed documentation, visit the following links:


Note: The Git protocol is not encrypted or authenticated, and only allows read access.

Start and enable git-daemon.socket.

The daemon is started with the following options:

ExecStart=-/usr/lib/git-core/git-daemon --inetd --export-all --base-path=/srv/git

Repositories placed in /srv/git/ will be recognized by the daemon. Clients can connect with something similar to:

$ git clone git://location/repository.git

Setting access rights

To restrict read and/or write access, use standard Unix permissions. Refer to http://sitaramc.github.com/gitolite/doc/overkill.html[dead link 2013-11-06] (archive.org mirror) for more information.

For fine-grained access management, refer to gitolite and gitosis.

See also